True, but it does affect the impact of Document Quality scores. Changing the 
range by an order of magnitude or more can wipe out any effect of DQ if you 
made assumptions about the possible range of values. Of course, you can 
compensate for various ranges if you know what they are, but this one caught us 
by surprise (guess we need to pay more attention to release notes).

Also, what do you mean by "some queries?" Are there rules for when a score 
falls into one range versus another?

-----Original Message-----
From: [email protected] 
[mailto:[email protected]] On Behalf Of Danny Sokolsky
Sent: Sunday, October 30, 2011 8:16 PM
To: General MarkLogic Developer Discussion
Subject: Re: [MarkLogic Dev General] Search scores have gone up dramatically

There was a bug fix in 4.2-5 that raised scores on some queries.  This was done 
to improve the precision on scores that are relatively low (as Mike surmised).  
Also, as Mike points out, it should not change the relevance from result to 
result in a given search, and scores between different searches are not 
comparable--they are for comparing between results in a given search.

-Danny
________________________________________
From: [email protected] 
[[email protected]] On Behalf Of Michael Blakeley 
[[email protected]]
Sent: Friday, October 28, 2011 12:37 PM
To: General MarkLogic Developer Discussion
Subject: Re: [MarkLogic Dev General] Search scores have gone up dramatically

Just for background, I'll mention that TF-IDF scores are relative to the 
contents of your database. So if the database changes, the scores for queries 
may also change. If your database is active, scores will almost certainly 
change over time. In general this is more noticeable with smaller databases, 
but it happens with any database.

That aside, if I noticed a change of several orders of magnitude, I too would 
suspect a change in algorithm. Without having any inside information, I suppose 
that it could be a bug, or the MarkLogic folks decided it would work better 
with more dynamic range. Scores are xs:integer, and there is a fair amount of 
math involved, so accuracy could be better using larger scores.

-- Mike

On 28 Oct 2011, at 10:39 , John Mulholland wrote:

> I have noticed that version 4.2-4 scores were lower than 1000.  As of version 
> 4.2-5 through 4.2-7 the scores have gone up dramatically, over 100000.  I 
> have been unable to find any information on this change.  Does anyone have 
> any insight?
>
> John
>
>
>
> NOTICE: This email message is for the sole use of the intended recipient(s) 
> and may contain confidential and privileged information. Any unauthorized 
> review, use, disclosure or distribution is prohibited. If you are not the 
> intended recipient, please contact the sender by reply email and destroy all 
> copies of the original message.
>
> _______________________________________________
> General mailing list
> [email protected]
> http://developer.marklogic.com/mailman/listinfo/general

_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general

Reply via email to